Introduction

When presenting the data overview and exploratory analysis results, we used to copy a lots tables, charts from Rstudio to PowerPoint, which makes the presentation preparation painful. It become essential for data scientist to make use of better reporting tools, such as R markdown, jupyter notebooks to author analysis presentation in a more efficient and organized way, of course, we also want this to be reproducible!

In this post, I would like to share some tips when I explore building analysis report using R markdown/notebook.

R markdown

Yihui Xie provided a very comprehensive and updated version of R markdown cookbook at https://bookdown.org/yihui/rmarkdown/, which gives good explanation how to make table of content, fold code snippets, configure the yaml header and more.

Tables

The native markdown table isn’t very user-friendly, so we have to make use of functions such as knitr::kable or DT::datatable to render the table from data.frame.

I would like to share some tips on choosing between kable and datatable.

• kable has simpler syntax and give more appealing “table like” tables in most themes.
• datatable has more capability such as paged tables with download buttons. There are more configurations could be referred from its JavaScript API specifications.

In a nutshell, kable is preferable for smaller tables, while datatable is preferable for bigger tables.

| Tables   |      Are      |  Cool |
|----------|:-------------:|------:|
| col 1 is |  left-aligned | $1600 |
| col 2 is |    centered   |   $12 |
| col 3 is | right-aligned |    $1 |

require(knitr)
require(kableExtra)
mtcars %>%
  head() %>%
  kable(digits = 1, caption = 'example of kable table') %>%
  kable_styling(full_width = FALSE, position = 'left') %>%
  row_spec(0,
           bold = T,
           color = 'white',
           background = 'black')

mtcars %>%
  datatable(
    class = 'cell-border stripe',
    extensions = 'Buttons',
    options = list(
      dom = 'Bfrtip',
      buttons = c('copy', 'csv', 'print')
    )
  )

Data Summary

summartools

require(summarytools)
mtcars %>%
  dfSummary(style = 'grid',
            graph.magnif = 0.75,
            plain.ascii = F,
            valid.col = FALSE,
            tmp.img.dir = "/tmp") %>%
  print()

Data Frame Summary

mtcars

Dimensions: 32 x 11
Duplicates: 0

No	Variable	Stats / Values	Freqs (% of Valid)	Graph	Missing
1	mpg [numeric]	Mean (sd) : 20.1 (6) min < med < max: 10.4 < 19.2 < 33.9 IQR (CV) : 7.4 (0.3)	25 distinct values		0 (0%)
2	cyl [numeric]	Mean (sd) : 6.2 (1.8) min < med < max: 4 < 6 < 8 IQR (CV) : 4 (0.3)	4 : 11 (34.4%) 6 : 7 (21.9%) 8 : 14 (43.8%)		0 (0%)
3	disp [numeric]	Mean (sd) : 230.7 (123.9) min < med < max: 71.1 < 196.3 < 472 IQR (CV) : 205.2 (0.5)	27 distinct values		0 (0%)
4	hp [numeric]	Mean (sd) : 146.7 (68.6) min < med < max: 52 < 123 < 335 IQR (CV) : 83.5 (0.5)	22 distinct values		0 (0%)
5	drat [numeric]	Mean (sd) : 3.6 (0.5) min < med < max: 2.8 < 3.7 < 4.9 IQR (CV) : 0.8 (0.1)	22 distinct values		0 (0%)
6	wt [numeric]	Mean (sd) : 3.2 (1) min < med < max: 1.5 < 3.3 < 5.4 IQR (CV) : 1 (0.3)	29 distinct values		0 (0%)
7	qsec [numeric]	Mean (sd) : 17.8 (1.8) min < med < max: 14.5 < 17.7 < 22.9 IQR (CV) : 2 (0.1)	30 distinct values		0 (0%)
8	vs [numeric]	Min : 0 Mean : 0.4 Max : 1	0 : 18 (56.2%) 1 : 14 (43.8%)		0 (0%)
9	am [numeric]	Min : 0 Mean : 0.4 Max : 1	0 : 19 (59.4%) 1 : 13 (40.6%)		0 (0%)
10	gear [numeric]	Mean (sd) : 3.7 (0.7) min < med < max: 3 < 4 < 5 IQR (CV) : 1 (0.2)	3 : 15 (46.9%) 4 : 12 (37.5%) 5 : 5 (15.6%)		0 (0%)
11	carb [numeric]	Mean (sd) : 2.8 (1.6) min < med < max: 1 < 2 < 8 IQR (CV) : 2 (0.6)	1 : 7 (21.9%) 2 : 10 (31.2%) 3 : 3 ( 9.4%) 4 : 10 (31.2%) 6 : 1 ( 3.1%) 8 : 1 ( 3.1%)		0 (0%)

Static Plots

ggplot2 is our best friend in R visualization and it has good support in R markdown. Chaining functions using %>% and + makes the code chunk beautiful!.

A lot of times, we would like combined many sub-plots into one. ggplot2::facet_grid could do some of jobs, but I found ggpubr::ggarrange is more powerful that allow you to combined any plots and even tables. It’s cool to put chart and table side by side. (example is given in subsequent section)

ggrigdes is another useful ggplot extension that plots multiple density plots in a single chart. This is often used when comparing profiles between groups. check the detail from here: https://cran.r-project.org/web/packages/ggridges/vignettes/introduction.html

require(ggplot2)
cor(mtcars) %>%
  as.data.frame() %>%
  tibble::rownames_to_column('var1')  %>%
  tidyr::pivot_longer(-var1, names_to = 'var2', values_to = 'cor') %>%
  filter(var1 <= var2) %>%
  ggplot(aes(x = var1, y = var2, fill = cor, label = round(cor,2))) +
  geom_tile() +
  geom_text() +
  scale_fill_gradient2() +
  labs(title = 'example of ggplot2 in R markdown')

require(ggpubr)
require(forcats)
# add .groups = 'drop' to remove some warnings from `dplyr`
data <- mtcars %>%
  group_by(gear) %>%
  summarise(n = n(), .groups = 'drop') %>%
  ungroup() %>%
  mutate(gear = fct_rev(factor(gear)))

plt <- data %>%
  ggplot(aes(x  = gear, y = n)) +
  geom_bar(stat = 'identity', fill = 'lightblue') +
  coord_flip() +
  labs(title = 'example of combined table and plot using ggpubr')

tbl <- ggtexttable(data, rows = NULL)
ggarrange(plt, tbl, ncol =2 , nrow = 1, widths = c(2,1))

Interactive Plots

This is the section that becomes tricky. Interaction plots are only supported in HTML R document and there is no dominating interactive visualization packages in R environment.

• plotly provides comprehensive chart types, documentation and cross-language capability. However I personally don’t like the style, syntax and toolbox at the right upper corner.

• echarts4r is a R interface for Echarts JavaScript library, which was open sourced by Baidu. I have tested the latest version 0.3.2 and it works well with R markdown.

• googleVis is a R interface for Google Charts JavaScript library, which was of course developed by Google. The package has a good collection of different chart types, but it has some unknown incompatibility with both R markdown and Shiny. I’ve found a workaround to integrate googleVis charts in R markdown, but it’s not perfect.

require(plotly)
p <- mtcars %>%
  mutate(am = factor(am)) %>%
  ggplot(aes(x = mpg, y = wt, color = am)) +
  geom_point() +
  labs(title = 'example of plotly in R markdown')
ggplotly(p)

library(echarts4r)
df <- data.frame(
  x = LETTERS[1:5],
  y = runif(5, 1, 5),
  z = runif(5, 3, 7)
)

df %>% 
  e_charts(x) %>% 
  e_radar(y, max = 7, name = "radar") %>%
  e_radar(z, max = 7, name = "chart") %>%
  e_tooltip(trigger = "item")

output: 
  html_document:
    self_contained: false

# self_contained: false is required for googleVis charts render in R markdown
suppressPackageStartupMessages(library(googleVis))
op <- options(gvis.plot.tag="chart")
plot(gvisHistogram(dino))